Architected a full-stack intelligence platform that evolved from a standard CRUD application into an autonomous AI agent. Designed a multi-step reasoning engine using Gemini 1.5 Flash and Fast API that can execute complex data mutations via natural language commands. Integrated a pgvector database for semantic search and Retrieval-Augmented Generation (RAG), allowing users to query unstructured data with high accuracy. Built robust fallback mechanisms (Mock Mode) to ensure continuous operation during API rate limits, all surfaced through a responsive React/Tailwind frontend.

Player Management: Create, Read, and Delete players.

Scouting Notes: Add detailed observations linked to specific players.

Tech Stack: React (Vite), Tailwind CSS, Fast API, PostgreSQL (AsyncSQLAlchemy).

Vector Search: Uses pgvector to store AI embeddings of notes.

Semantic Retrieval: Search for concepts like "defense" or "clutch" to find relevant notes even if exact keywords are missing.

Grounded Answers: The AI summarizes the search results and cites specific Note IDs as evidence.

Multi-Step Reasoning: The Agent can plan actions. If asked to "Add a note for a new player," it will:

Robustness: Includes a "Mock Mode" fallback if AI quotas are exceeded, ensuring the demo never crashes.

Transparency: The UI displays a live "Execution Log" showing the tool calls (Green steps).

Docker (for PostgreSQL + pgvector)

Python 3.10+

Node.js 18+

Google Gemini API Key

Hybrid AI Logic: Uses gemini-1.5-flash for high-speed inference but falls back to a Regex-based Mock Parser if rate limits are hit.

State Management: Chat history is lifted to the root component to persist across tab navigation.

Database Integrity: The backend explicitly handles async transaction commits to ensure the Agent "sees" data it just created within the same loop.